Robustness, answer quality, and efficiency beyond the controlled evaluation

Investigate cross-chunking robustness, end-to-end answer quality, and systems efficiency—including latency, memory use, and API cost—for NativeScope beyond the reported controlled retrieval evaluation.

Background

The reported correct-anchor evaluation uses only the default chunking and focuses on native-unit recall rather than generated-answer quality or systems performance. The paper also states that latency, memory, and API cost were not evaluated, while generative question answering was stopped after only 36 of 800 planned responses. These omissions leave unresolved whether NativeScope's observed retrieval benefits persist across chunking choices, translate into end-to-end answer-quality improvements, and justify its operational costs.